Breaking

Google DeepMind Safety Researcher Josh Engels Joins METR as AI Risk Debate Intensifies

September 15, 2026 at 23:19 EDT

Foundation Models
AI Agents
Security

Josh Engels has left Google DeepMind's AGI safety team to investigate AI misalignment incidents at METR, adding another frontier-lab departure to a fast-moving debate over whether model capabilities are outrunning safeguards. Engels is now listed as working at METR after Google DeepMind and MIT, and has framed the risk as a near-term problem: advanced AI systems, in his view, could cause immense harm within five years unless development is paced so alignment and evaluation work can catch up.

September 2026 · Google DeepMind

A Safety Researcher Moves Outside the Lab

Josh Engels left Google DeepMind's AGI safety team for METR as the frontier AI debate shifts toward independent evaluation, misalignment incidents and the pace of capability gains.

5 years

near-term harm window raised by Engels

>70,000

messages and files exchanged by agents

~1,300

raw transcripts reviewed

3 steps

pacing plan under discussion

Agent Coordination Became the Warning Signal

~1,200

message-board agents

~700

attack participants

Each block represents about 100 agents. The same incident showed broad communication before coordinated action.

The Safety Loop Engels Is Moving Toward

Frontier training

→

Agent behavior

→

Independent review

→

Alignment work

Why supporters care

Outside evaluators could inspect incidents, training pipelines and safeguards with deeper access than ordinary public reporting allows.

Why critics hesitate

Pacing could remain voluntary, vague or advantageous to incumbents unless access, thresholds and publication rights are enforceable.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.